The general way of making a Czech diphone voice is described in voice-czech-ph
package.  This document describes how these new Czech voices were processed.
This process was simplified because we received the recordings with diphone
boundaries already labeled.

You need to have an extracted source tarball of Festvox, the tarball is
available from http://www.festvox.org/download.html .

* Copying the recording

The received *.wav files were renamed appropriately and put into `wav'
subdirectory.  The received diphone index with timings was converted to
Festival + Festival Czech notation and put to dic/krbdiph.est.  It may be
necessary to invert the WAV files, e.g. using sox like:

  $ sox -v -1 INPUT.wav OUTPUT.wav

* Generating the #-# diphone (silence)

Sox options have changed since voice-czech-ph so we must use modify the sox
command line:

  $ dd if=/dev/zero of=krb0000.raw bs=100000 count=1
  $ sox -r 32k -e signed -b 16 -c 1 krb0000.raw krb0000.wav

* Looking for pitchmarks

Using the copied scripts from voice-czech-ph:

  $ export ESTDIR=/usr
  $ export FESTVOXDIR=...path-to-source-directory.../festvox
  $ mkdir pm
  $ ./tools/make_pm_wave wav/*.wav
  $ ./tools/make_pm_fix pm/*.pm

It may be necessary to change PM_ARGS in ./tools/make_pm_wave.

* Generating LPC parameters

tools/make_lpc script was copied from voice-czech-ph and -lpc_order parameter
was adjusted according to the recording sampling rate.  Then the script was
run:

  $ export ESTDIR=/usr
  $ export FESTVOXDIR=...path-to-source-directory.../festvox
  $ mkdir lpc
  $ ./tools/make_lpc wav/*.wav

* Creating the voice definition file

`festvox' directory was created.  The language definition file in it was copied
and renamed from its voice-czech-ph counterpart.  Everything in it was renamed
from `ph' to `krb'.  Additionally f0_mean was adjusted, to a common frequency
of the recording determined by examining some *.wav file using a sound editor.
f0_std was also adjusted in accordance with the f0_mean change.

* Creating Makefile

Makefile and make-group.scm files were copied from voice-czech-ph and adjusted.
Especially everything about power factors was removed from Makefile as we
hadn't adjusted power factors in this recording.
